Insertions Using Geometric Analysis and Hybrid Force-Position Control on a PUMA 560 with VAL II

David R. StripAAAI 1987

Abstract

This research presents a generalized methodology for robotic "peg-in-the-hole" insertions, moving beyond simple round pegs to complex, unchamfered shapes like triangles and stars. By combining hybrid force-position control with a multi-stage geometric strategy, the researchers enabled a PUMA 560 robot to autonomously align and insert a wide variety of parts using a single, flexible program.

Introduction

In the 1980s, robotic assembly was largely limited to highly structured environments. The classic "peg-in-the-hole" problem had been solved for round, chamfered (beveled) pegs using passive hardware like Remote Center Compliance (RCC) devices. However, these mechanical solutions were "dumb" — they couldn't adapt to complex geometries like squares, stars, or threaded parts, nor could they handle unchamfered holes where the slightest misalignment leads to a collision.

The authors sought to replace rigid, part-specific hardware with active compliance: a software-driven approach where the robot senses forces in real-time and adjusts its position to "feel" its way into a hole. The goal was to create a truly flexible manufacturing system where a robot could be programmed once to handle an entire tray of differently shaped parts.

Methodology

The researchers used a PUMA 560 manipulator equipped with a 6-axis force-torque sensing wrist. Their approach relies on two main pillars:

Hybrid Force-Position Control

Based on Mason’s earlier theoretical models, the authors partitioned the robot’s six degrees of freedom. For any given task, some axes are "position-controlled" (move to coordinate X), while others are "force-controlled" (push with Y Newtons). They implemented this via a custom command called MCOMPLY within a LISP-based environment (SCHEME), which communicated with the robot's VAL II controller.

Multi-Stage Geometric Strategy

Instead of trying to drop a peg straight in, they developed a multi-step process to manage initial contact and alignment:

  1. Approach: The peg is tilted and brought into contact with the surface, ensuring a specific "target point" (usually the sharpest corner) is over the hole.
  2. Orientation: The robot drives the target point into the corner of the hole. As it does so, the contact forces create torques that naturally rotate the peg into alignment.
  3. Insertion: Once aligned, the robot switches to a mode where it maintains a downward force while compliantly correcting for any binding or friction.
Four stages of the robotic insertion process: approach, chamfer crossing, one-point contact, and two-point contact.

Key Insights and Ideas

  • The "Target Point" Concept: The most elegant idea in the paper is using geometry to simplify control. By identifying the point with the smallest radius of curvature (the sharpest corner), the robot can induce predictable torques. This allows the physics of the contact to do the "math" of alignment.
  • Active vs. Passive Compliance: While RCC devices are fast, they are limited by their physical spring constants. Software-based active compliance allows the robot to act like a "programmable spring," changing its stiffness on the fly to suit different materials or shapes.
  • Guarded Moves: The implementation includes "guarded move" capabilities, where the robot automatically stops or changes behavior if a force threshold is exceeded, preventing damage to fragile parts.

Experimental Results

The system was tested on a physical suite of objects including round, square, rectangular, triangular, oval, and star-shaped pegs. Key findings include:

  • Versatility: The researchers proved that a single program could insert a variety of shapes that would typically require different mechanical jigs.
  • Precision via Sensing: The system could handle linear displacements of up to 0.25 inches and angular errors of 4-5 degrees from the vision system, correcting these errors through tactile feedback during the orientation stage.
  • Force Regulation: The system successfully handled fragile objects by maintaining constant, low-level contact forces, which is impossible with traditional high-stiffness position control.

Limitations and Future Work

  • Control Lag: The authors noted a significant "lag" (approx. 28ms) between sensing a force and the robot reacting. This latency, combined with the mechanical stiffness of the arm, caused the peg to occasionally "bounce" at the bottom of a hole rather than stopping cleanly.
  • Coordinate Frame Restrictions: The current implementation could only partition forces along the Cartesian axes of the tool frame. They noted that the ability to rotate the force-reference frame would be necessary for even more complex, non-orthogonal tasks.
  • Future Directions: The paper concludes by suggesting that the next step is automatic strategy generation—where a robot looks at a CAD description of a part and automatically determines the "target point" and insertion path without human intervention. They also mention moving toward "fine control" using multi-fingered hands rather than moving the entire massive robot arm for tiny adjustments.

Summary updated

BibTeX
@inproceedings{strip1987insertions,
  author = {David R. Strip},
  title = {{Insertions Using Geometric Analysis and Hybrid Force-Position Control on a PUMA 560 with VAL II}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1987},
}
Figure